Discrete Differential Geometry (600.657)misha/Fall09/1-curves.pdfDifferential Geometry of Curves and...
Transcript of Discrete Differential Geometry (600.657)misha/Fall09/1-curves.pdfDifferential Geometry of Curves and...
Differential Geometry:Curves, Tangents, and Curvature
Differential Geometry of Curves and Surfaces, Do Carmo (Chapter 1)
Differentiable Curves
Definition:
A parameterized differentiable curve is a differentiable map α:I→Rn of an open interval I=(a,b) of the real line R into Rn.
Differentiable Curves
Examples:
• α(u)=(cos u , sin u , u)
• α(u)=(u , |u|)
• α(u)=(u3-4u , u2-4)
• α(u)=(u3 , u2)
( )uuuu ,sin,cos)( =α
( )4,4)( 23 −−= uuuuα ( )23,)( uuu =α( )uuu ,)( =α
Differentiable Curves
Note:
If φ(u) is a differentiable map from [a’,b’] to [a,b], then the function αoφ:[a’,b’]→Rn is also a parameterized differentiable curve tracing out the same points.
( )4,4)( 23 −−= uuuuα ( )23,)( uuu =α
Regular Curves
Given a parameterized curve, α(u), the tangentof α at a point u is the vector defined by the derivative of the parameterization, α’(u).
( )23,)( uuu =α
Regular Curves
Given a parameterized curve, α(u), the tangentof α at a point u is the vector defined by the derivative of the parameterization, α’(u).
The parameterized curve α(u) is regular if α’(u)≠0 for all u∈I.
( )23,)( uuu =α
Regular Curves
Given a parameterized curve, α(u), the tangentof α at a point u is the vector defined by the derivative of the parameterization, α’(u).
The parameterized curve α(u) is regular if α’(u)≠0 for all u∈I. This implies:
–There is a “well-defined” tangent lineat each point on the curve.
–The map α(u) is “injective”
( )4,4)( 23 −−= uuuuα
Regular Curves
We say that two parameterized, differentiable, regular, curves α:[a,b]→Rn and β:[a’,b’]→Rn
define the same differentiable curve if there exists a differentiable map φ:[a’,b’]→[a,b], with φ’(u)≠0, such that βoφ=α.
Regular Curves
We say that two parameterized, differentiable, regular, curves α:[a,b]→Rn and β:[a’,b’]→Rn
define the same differentiable curve if there exists a differentiable map φ:[a’,b’]→[a,b], with φ’(u)≠0, such that βoφ=α.We say that they define the same oriented differentiable curve if, in addition, φ’(u)>0.
Regular Curves
We say that two parameterized, differentiable, regular, curves α:[a,b]→Rn and β:[a’,b’]→Rn
define the same differentiable curve if there exists a differentiable map φ:[a’,b’]→[a,b], with φ’(u)≠0, such that βoφ=α.We say that they define the same oriented differentiable curve if, in addition, φ’(u)>0.From now on, we will talk about (oriented) curves as equivalence classes, and parameterized curves as instances of a class.
Regular Curves
Given a parameterized curve α(v), and given u∈I, the arc-length from the point u0 is:
∫=u
u
dvvus0
)(')( α
Regular Curves
Given a parameterized curve α(v), and given u∈I, the arc-length from the point u0 is:
If we partition the interval [u0,u] into N sub-intervals, setting ∆u=(u-u0)/N and ui=u0+i∆u, we can approximate the integral as:
∫=u
u
dvvus0
)(')( α
( )
( ) ( )
( ) ( )uuuu
uu
uu
uuus
N
iiiN
N
i
ii
N
N
iiN
∆∆
−=
∆∆−
=
∆=
∑
∑
∑
−
=+∞→
−
=
+
∞→
−
=∞→
1
01
1
0
1
1
0
lim
lim
'lim)(
αα
αα
α
α(u0)
α(ui) α(uN)
Regular Curves
Using the arc-length, we can integrate functions defined over the curve.Given a function f:Rn→R, the integral of f over the curve α can be obtained by integrating over the parameterization domain I:
( )∫∫ =∈
b
ap
duuufdppf )(')()( ααα
Regular Curves
Using the arc-length, we can integrate functions defined over the curve.Given a function f:Rn→R, the integral of f over the curve α can be obtained by integrating over the parameterization domain I:
Note: If the parameterizations α and β define the same curve, they define the same integral. So integrals are properties of curves.
( )∫∫ =∈
b
ap
duuufdppf )(')()( ααα
Regular Curves
Given a regular differentiable curve α, we can define the unit-tangent vector at a point α(u) as:
)(')(')(
ttut
αα
=
Regular Curves
Given a regular differentiable curve α, we can define the unit-tangent vector at a point α(u) as:
Note: If the parameterizations α and β define the same oriented curve, they define the same unit-tangent vectors. So unit-tangent vectors are properties of oriented curves.
)(')(')(
ttut
αα
=
Regular Curves
Given a regular differentiable curve α, we can define the unit-tangent vector at a point α(u) as:
Note: If the parameterizations α and β define the same oriented curve, they define the same unit-tangent vectors. So unit-tangent vectors are properties of oriented curves.Tangent lines are properties of (un-oriented) curves.
)(')(')(
ttut
αα
=
Curvature
The tangent line at a point α(u0) on the curve describes the best-fitting line to the curve.
α(u0)
Curvature
The tangent line at a point α(u0) on the curve describes the best-fitting line to the curve.We would like to extend this to finding the best-fit circle at a point.
α(u0)
Curvature
Q: How can we compute the (reciprocal of the) radius of a circle using only local information?
Q: How can we compute the (reciprocal of the) radius of a circle using only local information?A: Look at how quickly the tangent vectors are turning as we move between adjacent points.
Curvature
ϕ ϕp1
p0 p0
p1
t1
t0
t1
t0
Curvature
Q: How can we compute the (reciprocal of the) radius of a circle using only local information?A: Look at how quickly the tangent vectors are turning as we move between adjacent points.We use the fact that the length of the chord between p0 and p1 equals the distancebetween the tangents, multipliedby the circle’s radius: ϕ
p0
p1
t1
t0
rt1rt0
0101 ppttr −=−
Curvature
Thus, we can express the radius as:
01
011pptt
r −−
=
ϕp0
p1
t1
t0
rt1rt0
Curvature
Thus, we can express the radius as:
In the case of the curve, we can compute the curvature by looking at the limit as u1 goes to u0:
01
011pptt
r −−
=
)(')('
)()()()(
lim)(0
0
01
010
01 uut
uuutut
uuu ααα
κ =−−
=→
Curvature (2D)
For the case of an oriented curve in 2D, we can assign a sign to the curvature, making it positive if the center of the circle is to the left of the tangent and negative if it is to the right.
0<κ
0>κ
Curvature (2D)
For the case of an oriented curve in 2D, we can assign a sign to the curvature, making it positive if the center of the circle is to the left of the tangent and negative if it is to the right.
Note: If parameterizations α and β definethe same (oriented) curve, theydefine the same (signed)curvature. So curvature isa property of curves.
0<κ
0>κ
Curvature (2D)
Theorem:For a closed oriented curve α⊂R2, the integral of the curvature over α is an integer multiple of 2π:
∫∈
=α
πκp
ldpp 2)(
Curvature (2D)
Rough Proof:To prove this, we show that the integral of the curvature is equal to the integral of the change of angle traced out by the tangent indicatrix.
Curvature (2D)
Definition:Given an oriented differentiable curve α:I→R2, we can define the tangent indicatrix, a map from the interval I to the unit-circle defined by the tangent.
Curvature (2D)
Definition:Given an oriented differentiable curve α:I→R2, we can define the tangent indicatrix, a map from the interval I to the unit-circle defined by the tangent.
Curvature (2D)
Definition:Given an oriented differentiable curve α:I→R2, we can define the tangent indicatrix, a map from the interval I to the unit-circle defined by the tangent.
Curvature (2D)
Definition:Given an oriented differentiable curve α:I→R2, we can define the tangent indicatrix, a map from the interval I to the unit-circle defined by the tangent.
Curvature (2D)
Definition:Given an oriented differentiable curve α:I→R2, we can define the tangent indicatrix, a map from the interval I to the unit-circle defined by the tangent.
Curvature (2D)
Rough Proof:1. Using the tangent indicatrix, we can assign an
angle to each point on the curve, (the angle of the tangent vector).
Curvature (2D)
Rough Proof:2. With a bit of manipulation, we can show that
the change in the angle, as a function of arc-length, is equal to the curvature.
Curvature (2D)
Rough Proof:2. With a bit of manipulation, we can show that
the change in the angle, as a function of arc-length, is equal to the curvature.
Recall that the (signed) curvature is measured as the (signed) change in tangent length, divided by the change in arc-length.
)(')('
)(0
00 u
utu
ακ ±=
Curvature (2D)
Rough Proof:2. With a bit of manipulation, we can show that
the change in the angle, as a function of arc-length, is equal to the curvature.
Recall that the (signed) curvature is measured as the (signed) change in tangent length, divided by the change in arc-length.For small step sizes, the change in tangent length is equal to the change in angle:
ϕϕ ≈sin
Curvature (2D)
Rough Proof:3. Since the integral is the accumulation of
changes of angle, weighted by arc-length, and since we start and end with the same angle, the accumulated angle must be an integer multiple of 2π.
Curvature (2D)
Theorem:For a closed curve α⊂R2, the integral of the curvature over α is an integer multiple of 2π:
The integer l is the winding index of the curve.∫∈
=α
πκp
ldpp 2)(
Curvature (2D)
In defining the curvature, we implicitly used two different definitions:1. Curvature is the (signed) distance between
tangents divided by the distance between the corresponding points on the circle.
ϕp0
p1
t1
t0
01
010
01
lim)(pptt
ppp −
−=
→κ
Curvature (2D)
In defining the curvature, we implicitly used two different definitions:2. Curvature is the (signed) angle between
tangents divided by the arc-length between the corresponding points on the circle.
ϕp0=α(u0)
t1
t0
ϕ
∫→
=1
0
01
)('lim)( 0 u
u
uudvv
pα
ϕκp1=α(u1)
Curvature (2D)
In defining the curvature, we implicitly used two different definitions.In the limit, the two definitions are the same since the ratio of chord-length to arc-length goes to one as points get closer:
ϕ
)(22
sin2 3ϕϕϕ O+= sin(ϕ/2)ϕ/2
Curvature (2D)
In defining the curvature, we implicitly used two different definitions.In the limit, the two definitions are the same since the ratio of chord-length to arc-length goes to one as points get closer.But in the discretesetting, they givesrise to two differentnotions of curvature.
ϕ
sin(ϕ/2)ϕ/2
Discrete Curves
Definition:
A discrete curve is the set of points on the line segments between consecutive vertices in an ordered set of vertices {v0,…,vN}⊂Rn, such that for any index i, vi≠vi+1. vi
vi+1
Discrete Curves
The length of the discrete curve between vertexvi0 and vi is:
∑−
=+ −=
1
10
i
ijjji vvl
vivi+1
Discrete Curves
The length of the discrete curve between vertexvi0 and vi is:
Given a function assigning a value fi to edge vivi+1, we can compute theintegral of the function as:
∑−
=+ −=
1
10
i
ijjji vvl
vivi+1
∑−
=+ −=
1
01
N
iiii fvvF
Discrete Curves
The length of the discrete curve between vertexvi0 and vi is:
Similarly, given a function assigning a value fi to vertex vi, we can compute theintegral of the function as:
∑−
=+ −=
1
10
i
ijjji vvl
vivi+1
∑−
=
−+ −+−=
1
0
11
2
N
ii
iiii fvvvv
F
Discrete Curves
At any (point on an) edge vivi+1, we can define the tangent vector ti as the unit vector pointing from vi to vi+1:
ii
iii vv
vvt−−
=+
+
1
1
vivi+1
ti
ti+1ti-1
Discrete Curves
Q: Where and how should we define curvature?
vivi+1
ti
ti+1ti-1
Discrete Curves
Q: Where and how should we define curvature?
A: Since the tangent only changes at vertices we should define the curvature as a vertex value.
vivi+1
ti
ti+1ti-1
Discrete Curves
Q: Where and how should we define curvature?
A: Since the tangent only changes at vertices we should define the curvature as a vertex value.
We should define the value of the curvature as the change in the tangent aswe move through the vertex.
vivi+1
ti
ti+1ti-1
Discrete Curves
Note that we cannot define the tangent as the result of a limiting process:
vivi+1
[ ] [ ]
11
1
0
11
1
0
lim
)1()1(lim
−+
−
→∆
−+
−
→∆
−∆−
=
∆+∆−−∆+∆−−
=
ii
ii
t
iiii
ii
ti
vvttt
tvvttvvttt
κ
ti
ti+1ti-1
Discrete Curves
However, we can estimate it using the finite-differences using the edge centers (vi+vi+1)/2.
vivi+1
ti
ti+1ti-1
21 ii vv ++
Discrete Curves
However, we can estimate it using the finite-differences using the edge centers (vi+vi+1)/2.
1. Define the curvature in terms of change in tangents divided by the distance between edge centers: vi
vi+1
ti
ti+1ti-1
21 ii vv ++
[ ] [ ]2/)(2/)( 11
1
iiii
iii vvvv
tt+−+
−=
−+
−κ
ti
ti-1
21 ii vv ++
21−+ ii vv
Discrete Curves
However, we can estimate it using the finite-differences using the edge centers (vi+vi+1)/2.
2. Define the curvature in terms of change in tangents angle divided by the arc-length between edge centers: vi
vi+1
ti
ti+1ti-1
21 ii vv ++
2/)(2/)( 11
21
−+ −+−∠
=iiii
i vvvvttκ
ti
ti-1
21 ii vv ++
21−+ ii vv
Discrete Curves
In the limit, as we refine the tessellation of a differentiable curve, both methods define the same curvature.
Does it make a difference which one we use?
Discrete Curves
Does it make a difference which one we use?
Consider integrating the curvature over the closed curve:
∑−
=
−+ −+−=Κ
1
0
11
2
N
ii
iiii vvvvκ
vivi+1
ti
ti+1ti-1
Discrete Curves
Does it make a difference which one we use?
Consider integrating the curvature over the closed curve:
Using the second definition,we get:
∑−
=
−+ −+−=Κ
1
0
11
2
N
ii
iiii vvvvκ
∑−
=
∠=Κ1
021
N
itt
2/)(2/)( 11
21
−+ −+−∠
=iiii
i vvvvttκ
vivi+1
ti
ti+1ti-1
21 ii vv ++
Discrete Curves
Does it make a difference which one we use?
Consider integrating the curvature over the closed curve:
Using the second definition,we get the sum of angulardeficits over the vertices ofa closed polygon.
∑−
=
−+ −+−=Κ
1
0
11
2
N
ii
iiii vvvvκ
vivi+1
ti
ti+1ti-1
21 ii vv ++
Discrete Curves
Does it make a difference which one we use?
Consider integrating the curvature over the closed curve:
Using the second definition,we get the sum of angulardeficits over the vertices ofa closed polygon.
This has to be a multiple of 2π sothis definition conforms to the winding index.
∑−
=
−+ −+−=Κ
1
0
11
2
N
ii
iiii vvvvκ
vivi+1
ti
ti+1ti-1
21 ii vv ++
Discrete Curves
Does it make a difference which one we use?
Consider integrating the curvature over the closed curve:
Using the first definition wewould get a value for thecurvature integral that isnot a multiple of 2π, so thisdefinition does not conform tothe winding index.
∑−
=
−+ −+−=Κ
1
0
11
2
N
ii
iiii vvvvκ
vivi+1
ti
ti+1ti-1
21 ii vv ++